Traffic mode selection prediction method based on computational graph

Abstract

The invention discloses a traffic trip mode selection prediction method based on a computational graph. According to the method, aiming at the prediction problem of transportation mode selection, a multi-term Logit model in a discrete selection model and an artificial neural network (ANN) of deep learning domain knowledge are combined through a specific Softmax activation function, and a set of calculation graph solving framework is created for calculation solving. In the computational graph framework, back propagation and stochastic gradient descent algorithms are combined to minimize the system loss error and improve the precision of the solving method. According to the method, a computational graph framework is applied to map a multi-term Logit model (MNL, multi-term Logit), forward transmission and reverse propagation can be accurately and efficiently carried out based on data, hidden transportation modes in a deep learning network can be better recognized and explained, and the efficiency of large-scale data solution calculation and prediction of transportation mode selection can be improved.

Description

[0001] Field [0002] The invention belongs to the field of traffic planning in transportation planning and management, and in particular relates to a method for selecting and predicting traffic travel modes based on calculation graphs. Background technique [0003] The prediction of traffic travel mode selection is an important content of urban traffic demand forecasting, and it is also an important basis for urban traffic planning and management. Analyzing the difference in travelers' choice of service attributes can provide a theoretical basis for improving the service level of urban transportation. In the classic traffic four-stage method, the prediction of urban traffic mode selection often adopts discrete choice models, the most important ones are Logit family models, such as multinomial Logit model (MNL, multinomial Logit), NL model (nested Logit) and Mixed Logit model, etc. The Logit model has a theoretical basis for mathematical proof, but requires input data to mee...

AI Technical Summary

Problems solved by technology

In addition, the prediction accuracy and interpretability of the current transportation mode selection prediction methods are insufficient. It takes a lot of cost to identify the transportation mode, and there is no perfect and mature system to completely and efficiently predict the traveler's transportation mode. , which brings great challenges to urban traffic planning and management

Images